Evaluation of Genetic Isolation Within an Island Flora Reveals Unusually

Home , Local adaptation

Evaluation of genetic isolation within an island flora reveals unusually widespread local adaptation and supports sympatric speciation rstb.royalsocietypublishing.org Alexander S. T. Papadopulos1, Maria Kaye2,Ce´line Devaux3, Helen Hipperson1, Jackie Lighten4, Luke T. Dunning1, Ian Hutton5, William J. Baker6, Roger K. Butlin7 and Vincent Savolainen1,6

Research 1Grand Challenges in Ecosystem and the Environment Initiative, Imperial College London, Silwood Park Campus, Ascot, Berkshire SL5 7PY, UK Cite this article: Papadopulos AST et al. 2014 2School of Biological Sciences, University of Aberdeen, Aberdeen AB24 3FX, UK 3 Evaluation of genetic isolation within an island Institut des Sciences de l’Evolution de Montpellier, UMR 5554, 34095 Montpellier, France 4Department of Biology, Dalhousie University, Halifax, Nova Scotia, Canada B3H 4R2 flora reveals unusually widespread local 5Lord Howe Island Museum, Lord Howe Island, PO Box 157, New South Wales 2898, Australia adaptation and supports sympatric speciation. 6Royal Botanic Gardens, Kew, Richmond, Surrey TW9 3DS, UK Phil. Trans. R. Soc. B 369: 20130342. 7Department of Animal and Plant Sciences, University of Sheffield, Sheffield S10 2TN, UK http://dx.doi.org/10.1098/rstb.2013.0342 It is now recognized that speciation can proceed even when divergent natural selection is opposed by gene flow. Understanding the extent to which environ- One contribution of 14 to a Theme Issue mental gradients and geographical distance can limit gene flow within species ‘Contemporary and future studies in plant can shed light on the relative roles of selection and dispersal limitation during speciation, morphological/floral evolution and the early stages of population divergence and speciation. On the remote Lord polyploidy: honouring the scientific contri- Howe Island (Australia), ecological speciation with gene flow is thought to butions of Leslie D. Gottlieb to plant have taken place in several plant genera. The aim of this study was to establish the contributions of isolation by environment (IBE) and isolation by com- evolutionary biology’. munity (IBC) to the genetic structure of 19 plant species, from a number of distantly related families, which have been subjected to similar environmental Subject Areas: pressures over comparable time scales. We applied an individual-based, multi- evolution, genetics, ecology variate, model averaging approach to quantify IBE and IBC, while controlling for isolation by distance (IBD). Our analyses demonstrated that all species experienced some degree of ecologically driven isolation, whereas only 12 of Keywords: 19 species were subjected to IBD. The prevalence of IBE within these plant local adaptation, ecological speciation, species indicates that divergent selection in plants frequently produces local genetic structure adaptation and supports hypotheses that ecological divergence can drive speciation in sympatry. Authors for correspondence: Alexander S. T. Papadopulos e-mail: [email protected] 1. Introduction Vincent Savolainen The role of natural selection in speciation has received renewed interest owing to e-mail: [email protected] the growing body of data that has demonstrated the potential for divergent selection to overcome the homogenizing effects of gene flow between populations [1–6]. A number of classic examples of ecological speciation—where the evolution of reproductive isolation between populations ultimately stems from divergent selection—have emerged [1,2,7]. However, there is still debate as to the extent to which ecologically driven isolation (e.g. via selection against hybrids and migrants) or geographically driven isolation (as a result of dispersal limitation) is the most significant component of speciation [8–10]. The tendency for geographically separated populations and individuals to be less likely to repro- duce with each other is manifested as a pattern of isolation by distance (IBD; Electronic supplementary material is available [11]). IBD is often quantified as the correlation between increasing neutral genetic at http://dx.doi.org/10.1098/rstb.2013.0342 or divergence and increasing geographical distance [12]. Alternatively, during via http://rstb.royalsocietypublishing.org.

& 2014 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/3.0/, which permits unrestricted use, provided the original author and source are credited. ecological speciation, environmental and ecological differences that have been subjected to broadly similar environmental fac- 2 between populations can constrain reproduction and migration. tors on a geologically similar time scale [8,26–28]. The product rstb.royalsocietypublishing.org This can occur as a result of selection against maladapted of a volcanic eruption 6.9 Ma, the tiny LHI is located 600 km individuals and alleles, which, in turn, reduces gene flow. This from the nearest land mass (Australia). The island is highly het- process may lead to local adaptation to particular environ- erogeneous with a mixture of geological formations (both ments and eventually to speciation [1,6,13]. Selection against basaltic and calcarenite) and a range of habitat types, ranging migrants and hybrids can yield a pattern similar to that from sclerophyllous temperate rainforest in the lowlands to stemming from IBD, where increasing environmental dissimi- moist cloud forest at the summit of Mt Gower (875 m), larity between locations is correlated with neutral genetic making it home to 90 endemic species [8,27,29]. Studying divergence between populations or individuals [14]. This pat- IBD, IBE and IBC in trees and bushes has a number of advan- tern of isolation by environment (IBE) is increasingly seen as a tages, including their static lifestyles, high dispersal abilities signature or precursor to incipient ecological speciation [13,15]. and propensity for local adaptation [20], and is particularly rel- B Soc. R. Trans. Phil. In reality, geographical and ecological processes that can evant to the flora of LHI. Recent studies have shown that influence spatial genetic patterns are not mutually exclusive, ecological speciation with gene flow is likely to have taken and decreased migration and reproduction can stem from place in multiple genera, and as much as 8.2% of the flora may both sources simultaneously. Thus, attempts to establish the have been the product of speciation that has taken place within contributions of geographical and ecological mechanisms to the confines of the island [8,10,26,28,30]. Previous research isolation in natural populations are complicated by the need has provided evidence that local adaptation has driven diver- 369 to disentangle the effects from one another [13–18]. Isolation sification in several of these genera (including Coprosma, 20130342 : driven by competition with other species in the local commu- Metrosideros and Howea [26,28]). IBD and IBE had weak but sig- nity may also be an important component of population nificant effects in several species, but only altitude and soil pH divergence. Such isolation by community (IBC) is often neg- were tested as explanatory variables of genetic variation [26,30]. lected in studies of IBE, which concentrate on the abiotic It is unclear whether the speciation and adaptation seen in environment, but it is a powerful driver of ecological displace- Coprosma, Metrosideros and Howea is peculiar to these genera, ment [19] and potentially of local adaptation. Additionally, or whether IBE is common throughout the LHI flora. The goal patterns of gene flow can arise where migration is not depen- of this study was to quantify the contributions of IBD, IBE and dent on the measured environmental gradient, but the IBC to genetic relatedness in 19 species. These include species movement of migrants is determined by additional factors, that are thought to be the products of sympatric speciation such as wind direction [18]. This can lead to counter-gradient events as well as those that have colonized LHI and evolved isolation where gene flow is higher between dissimilar allopatrically from their parent populations [8]. We also assess environments, further complicating observed patterns. whether IBD, IBE and IBC have led to the emergence of distinct Manystudies have examined only IBD or IBE, but a growing genetic clusters within any of the endemic LHI species, which number of different approaches have been exploited in order to may represent the early stages of the speciation continuum. quantify IBD and IBE at the same time (see [18,20] for reviews of these approaches). Two recent studies have examined patterns of IBD and IBE in a wide range of taxa [15,18], revealing 2. Material and methods that IBE may be a common phenomenon. A mixed effects meta-analysis of 106 studies found that environmental differ- (a) Study species and data collection ences accounted for 1.3–3.7% of neutral genetic divergence Genetic and ecological data were collected for individuals from when IBD was taken into account and 2.4–4.3% when it was 19 species for this study; 18 are endemic to LHI and one is a not [15]. A different study found that 52 of 70 studies provided non-endemic, native species (table 1). Two species (Coprosma sp. evidence of IBE or both IBE and IBD [18]. Frequently, studies nov. and C. putida-S) have not been formally described, but both have assessed IBE for one or a few environmental gradients, are genetically and morphologically distinct from other Coprosma or have summarized contribution of many environmental vari- populations [8,26] and are treated as discrete species here. Previous ables into a single measure of environmental distance [16]. research [8,26,27] suggests that nine of the endemic species (from the genera Coprosma, Metrosideros and Howea) are the products of Although Mantel tests and AMOVAs have been a mainstay speciation with gene flow that has occurred on LHI (referred of landscape genetics and assessments of IBD, IBE and isola- to as the ‘sympatric speciation group’, SSG). These within-island tion by adaptation [13,15–18,21–23], multivariate approaches speciation events can be considered as sympatric under the bio- (e.g. multiple matrix regression [16]) have received increas- geographic definition that we use here; however, under strict ing interest as they hold great potential for determining how population genetic definitions, these events may be considered geographical distance, phenotypic dissimilarity and multiple as parapatric speciation [26,30,31]. Population genetic markers landscape barriers can interact to alter patterns of relatedness (amplified fragment length polymorphisms, AFLPs [32]) and eco- [16–18,20]. Here, we exploit an extension of the multiple logical data have been generated for the SSG species previously matrix regression approach [16] by including a model averaging [26], and these data were combined with new data for a further procedure [24]. Model averaging provides the advantage that 10 species for this study. These data are composed of AFLP geno- uncertainty in parameter estimates due to model selection can types and environmental data for 10 variables collected in the field or extracted from GIS layers for each individual specimen (see [26] be taken into account, reducing the risk of false-positives for materials and methods). [24,25]. This approach also permits comparison of analyses The remaining nine endemic taxa are most likely to have that use a single, unified measure of environmental dissimilarity evolved anagenetically following colonization of the island [8]. with regressions of multiple variables. The LHI population of the non-endemic species Alyxia ruscifolia Lord Howe Island (LHI) provides a unique opportunity to differs in leaf morphology from the Australian population, but determine the prevalence of IBE in a single location across a it has not been described as a distinct subspecies [33]. These 10 range of phylogenetically closely and distantly related taxa species are subsequently referred to as the ‘allopatric speciation Table 1. Characteristics of study species. 3 rstb.royalsocietypublishing.org speciation seed typical genus species group endemic dispersal pollination habit n individuals n loci

Alyxia lindii allopatric yes animal animal scrambling 29 82 climber Alyxia ruscifolia allopatric no animal animal shrub 31 60 Atractocarpus stipularis allopatric yes animal animal small tree 46 223 Coprosma prisca allopatric yes animal wind shrub 49 104 Dracophyllum ﬁtzgeraldii allopatric yes wind animal tree 28 144 B Soc. R. Trans. Phil. Geniostoma petiolosum allopatric yes animal animal small tree 31 67 Macropiper excelsum subsp. allopatric yes animal wind shrub 33 175 psittacorum Macropiper hooglandii allopatric yes animal wind shrub 27 147 369 Xylosma maidenii allopatric yes animal wind small tree 41 164 20130342 : Zygogynum howeanum allopatric yes animal animal small tree 43 394 Coprosma huttoniana sympatric yes animal wind shrub to 43 819 small tree Coprosma lanceolaris sympatric yes animal wind shrub 102 819 Coprosma putida-N sympatric yes animal wind shrub to 111 819 small tree Coprosma putida-S sympatric yes animal wind small tree 26 819 Coprosma sp.nov sympatric yes animal wind scrambling 15 819 shrub Howea belmoreana sympatric yes wind animal tree 161 900 Howea forsteriana sympatric yes wind animal tree 188 900 Metrosideros nervulosa sympatric yes animal wind shrub to 78 478 small tree Metrosideros sclerocarpa sympatric yes animal wind tree 72 478 group’ (ASG). General features of the ecology of these species are combinations were carried out on five individuals from each listed in table 1. For DNA analyses, leaf tissue was collected from species. Three or four combinations for each species were mature individuals of these species on LHI and dried using silica chosen for the full analysis, based on the number of fragments gel. For consistency with the previous study [26], the altitude and and polymorphic loci (electronic supplementary material, appen- geographical position of each sample were recorded using an dix S1), with the exception of Coprosma prisca which was eTrex summit HC GPS with a built-in barometric altimeter. For genotyped with the same primer combinations as its congeneric the remaining nine variables, data for each specimen were species (see [26]). The raw data were analysed with GENEMAPPER extracted from the 10 Â 10 m raster grids of Papadopulos et al. V4 software (Applied Biosystems). Loci (bins) were defined by [26] based on GPS location. Variables included were; Euclidean eye in the range of 50–500 base pairs. Presence/absence at distance to the nearest creek, Euclidean distance to the coast each locus was scored automatically by GENEMAPPER, and scoring (a proxy for salt deposition), available light, soil water, soil pH, was subsequently confirmed manually. Only fragments with aspect of the slope, gradient of the slope and vector ruggedness signal intensity greater than 50 relative fluorescence units were (a measure of topographic heterogeneity [34]). The predominant scored as present. Samples were processed blindly, using extrac- winds on LHI come from the northeast [29]. To reflect this, aspect tion codes, to avoid subjectivity in peak scoring. Five individuals of the slope was converted into a continuous measure of north- of each species were selected at random, re-extracted and geno- easterly wind exposure ranging from 0 (equivalent to a typed. For these individuals, the two replicates were compared southwest aspect) to 180 (i.e. a northeast aspect) [26]. to identify mismatch errors between the genotypes. Loci with more than one mismatch error across the replicates were removed from further analyses. (b) Genotyping A method modified from the ‘CTAB’ protocol [35] was used to extract total genomic DNA from 0.3 to 0.5 g of dried leaf material (c) Multi-model inference of isolation by distance, [36]. DNA was purified using DNeasy mini spin columns isolation by environment and isolation (Qiagen, Crawley, West Sussex, UK) according to the manufac- turer’s protocol and subsequently quantified using a Nanodrop by community (ThermoScientific, Denver, CO). AFLP profiles were generated We applied a model averaging approach to determine potential for each sample as in [26]. Primer trials for 28 primer causes of genetic isolation between individuals within each species. This is an extension of the multiple matrix regression approach of collinearity between environmental variables was assessed by 4 Wang [16]. In this procedure, linear regression of a response distance calculation of variance inflation factors using the vif function in matrix (here, genetic relatedness) on two or more explanatory dis- the car package. For each species, variables with variance inflation rstb.royalsocietypublishing.org tance matrices is performed, while controlling interactions among factors of greater than five were removed from the analysis [41]. predictor variables. The correlation of each explanatory variable with the response variable was evaluated using model averaging, as described in Burnham & Anderson [24] and implemented by (d) Analysis of population structure the MuMIn package in R [37]. For each species, parameter estimates To determine whether distinct genetic clusters were present in for submodels (comprising all possible combinations of the included each species, the presence/absence AFLP data for the ASG predictor variables) were calculated. Coefficients (effect sizes), species were analysed using the individual-based Bayesian clus- unconditional standard errors and 95% confidence intervals for tering approach implemented in STRUCTURE v. 2.3.3 [42], adapted each predictor were calculated by averaging the estimates from sub- for use with dominant markers [43]. STRUCTURE analyses for the models in which each term appears and weighting values according SSG species have been performed previously [26]. For this B Soc. R. Trans. Phil. to the submodels’ Akaike information criterion (AICc) [24]. This study, all analyses were run using the admixture model with cor- approach has the added benefit that the uncertainty in parameter related allele frequencies and no a priori information of species/ estimates is known and it avoids pitfalls associated with model population membership. After preliminary runs, analyses of each selection and statistical null hypothesis testing [24,25]. Parameter dataset were conducted with K ¼ 1–8 clusters. For each value of estimates were considered significant when the 95% confidence K, 10 replicates of 80 000 Markov chain Monte Carlo iterations interval did not span zero [25]. In the context of this study, negative were run, and the first 10 000 iterations of each chain were dis- 369 effects of explanatory variables with kinship indicate that increasing carded. Two assessments were used to infer the number of geographical, environmental or community dissimilarity is corre- genetic clusters: (i) a comparison of the log probability of the 20130342 : lated with decreasing genetic relatedness. A positive effect denotes data (X ) given K Ln[Pr(X|K)] for different values of K; and a counter-gradient correlation, i.e. individuals are more genetically (ii) DK, the second-order rate of change in Ln[Pr(X|K)] [44]. related in dissimilar environments. To account for the distance Each species was analysed separately, with the exception of the matrix nature of our data, the analyses were repeated with 1000 per- two Macropiper species which were analysed together to detect mutations of the kinship matrix. The z-scores for variable coefficient interspecific hybridization between these close relatives. estimates from the original analysis were calculated and compared with those stemming from the permutation analyses to ensure that the patterns observed were not random. The resulting p-values 3. Results were used to control the false-discovery rate to 0.05 using fdrtool [38]. For each species, a genetic distance matrix composed of pair- (a) Three matrix analyses wise kinship coefficients was calculated using SPAGEDI 1.3 [39] When environmental dissimilarity was amalgamated into a assuming Hardy–Weinberg equilibrium. Within each species, single distance measure, significant IBD, IBE and/or IBC latitudinal and longitudinal coordinates for each individual were converted into a geographical distance matrix using aflpdat [40]. were detected in 17 species, but no effects were present in This method was preferred over the use of least cost path distances, Zygogynum howeanum and Coproma putida-S (figure 1). The which are increasingly used in animal studies [16,17], as there analysis revealed significant IBD in 12 species (six from each is no available information for the LHI plants to suggest that dif- group), IBE in 13 species (eight ASG versus six SSG) and IBC ferent habitats incur different dispersal limitations for pollen or in six species (two ASG versus four SSG). IBD was the sole diaspores. Unless otherwise stated, all further analyses were driver of genetic structure in one species (A. ruscifolia), as was performed in R [37], distance measures were calculated using the IBC (in Coprosma huttoniana). IBE was the only effect observed vegdist function in the vegan package. To construct a measure in three species. Both IBD and IBE were detected in nine of the difference in local community composition between individ- species, and three of these species were subjected to all three uals, we extracted the vegetative association [29] of each individual modes of isolation. C. prisca, was also subjected to IBD and from a 10 Â 10 m raster grid. Pickard described the most common counter-gradient effects of environmental dissimilarity. species present in each of his vegetative associations [29]. Using this information, we constructed a presence/absence matrix for each genus describing which species were likely to co-occur with (b) Twelve matrix analyses each specimen. This was then converted into a pairwise Jaccard The 12 matrix analyses, with each environmental variable dissimilarity matrix to describe the distance between individuals included as a separate predictor, produced results broadly con- in the composition of their local community. We evaluated the environmental variables in two ways. First, we performed a prin- sistent with the three matrix analyses. However, there were cipal component analysis using the prcomp function. The resulting important differences in some species, and IBE was more scores for each specimen we used to calculate a pairwise Euclidean widespread (figures 2 and 3). Variance inflation factors of distance matrix describing environmental dissimilarity between greater than five were detected in four species, and the affected specimens. Second, we calculated Euclidean distance matrices variables were removed from analysis in these species for each environmental variable separately. (C. prisca—altitude and soil water, Coprsma sp. nov.—proximity Two sets of model averaging analyses were performed for each to creeks and the coast, Macropiper excelsum—altitude, and species to determine the effect of combining environmental vari- Macropiper hooglandii—wind exposure). Isolating effects were ables into a single distance measure (as in [16]) versus assessing detected in 18 species, but not in Dracophyllum fitzgeraldii.Out the specific effects of each variable: (i) a three matrix analysis was of these 18 species, IBD was evident in 11 (six ASG versus performed using geographical distance (IBD), community four SSG), IBE (isolation by at least one environmental variable) dissimilarity (IBC) and the combined environmental dissimilarity matrices (IBE) as explanatory variables, with kinship as the was present in 17 species and IBC in four species (all SSG). response variable; and (ii) a 12 matrix analysis was performed Geographical distance was the sole driver of isolation only in using geographical distance (IBD), community dissimilarity (IBC) A. ruscifolia, whereas this was true for environmental variables and all 10 of the environmental dissimilarity matrices (IBE) as explain five species (three ASG versus two SSG). Significant counter- natory variables. Variables were standardized prior to analysis, and gradient effects were more common in these analysis; a positive Al. lind. Al. rusc. At. stip. Co. pris. Dr. fitz. Ge. peti. Ma. exce. Ma. hoog. Xy. maid. Zy. howe. 5 0.8 rstb.royalsocietypublishing.org 0.6 * **** ****** ** * ** 0.4 0.2 0 –0.2 –0.4 –0.6

–0.8 B Soc. R. Trans. Phil.

Co. hutt. Co. lanc. Co. put-N Co. put-S Co. sp. nv. Ho. belm. Ho. fors. Me. nerv. Me. scle. 0.8 12 0.6 ***** ********* ** 10 0.4 369

0.2 8 20130342 :

0 6 –0.2 4 –0.4 –0.6 2

–0.8 0 DEC DEC DEC DEC DEC DEC DEC DEC DEC DEC Figure 1. Three matrix model averaging results of IBD (D), IBE (E) and IBC (C) for 19 plant species on LHI. Each plot depicts effect sizes (points) and 95% confidence intervals (error bars) for each parameter. Effect sizes below zero denote negative correlations between predictor variables and relatedness. Effect sizes above zero indicate counter-gradient effects. Estimates were determined as significant (asterisk) when the 95% CI did not span zero and by permutation tests (a ¼ 0.045, FDR ¼ 0.05). The bar plot indicates the number of negative effects detected for each parameter across all species. ASG, filled circles, SSG, open circles. Al. lind., Alyxia lindii; Al. rusc., Alyxia ruscifolia; At. stip., Atractocarpus stipularis; Co. prisc., Coprosma prisca; Dr. fitz., Dracopyllum fitzgeraldii; Ge. peti., Geniostoma petiolosum; Ma. exce., Macropiper excelsum subsp. psittacorum; Ma. hoog., Macropiper hooglandii; Xy. maid., Xylosma maidenii; Zy. howe., Zygo- gynum howeanum; Co. hutt., Coprosma huttoniana; Co. lanc, Coprosma lanceolaris; Co. put-N, Coprosma putida-N; Co. put-S, Coprosma putida-S; Co. sp. nov., Coprosma sp. nov.; Ho. belm., Howea belmoreana; Ho. fors., Howea forsteriana; Me. nerv., Metrosideros nervulosa; Me. scle., Metrosideros sclerocarpa. correlation with geography in one species (C. huttoniana), analysed the ASG and found no clear population subdivision and with at least one environmental variable in seven species. in A. ruscifolia, Z. howeanum, M. excelsum subsp. psittacorum Differences between individuals in the distance to the nearest and M. hooglandii, although the analyses detected considerable creek was the most common source of isolation owing to hybridization between the two closely related Macropiper any individual variable (present in eight species), followed species. Statistical assessments indicate that there are two gen- by differences in proximity to the coast (six species). Counter- etic clusters in two widespread endemic species (electronic gradient effects due to elevation were the most common supplementary material, appendix S2): Atractocarpus stipularis (three species). and Xylosma maidenii (electronic supplementary material, The patterns of IBD, IBE and IBC are consistent between the table S1). However, the structure plot does not show clear sub- analyses in 11 of the 19 study species. When environmental division in Atractocarpus stipularis (figure 4). Based on DK, variables were distilled into a single measure of environmen- Alyxia lindii, C. prisca, D. fitzgeraldii and Geniostoma petiolosum tal dissimilarity, the pronounced differences occurred in: may have two populations present on the island, however C. huttoniana—IBC was gained while strong IBE (altitude the results were not conclusive as only a marginal increase in and creek) and counter-gradient effects of environment and [Pr(X|K)] for K ¼ 1overK ¼ 2 was evident. In X. maidenii, geography were lost; D. fitzgeraldii—no significant correlations the populations show some spatial separation along the were replaced with marginally significant effects of IBD island’s latitudinal gradient, but there was no clear spatial and IBC; Coprosma sp. nov.—IBE (slope) was replaced by separation of populations within the other species (figure 4). IBD; C. prisca, C. putida-S and Z. howeanum—the IBE effects (coast, wind and soil water, respectively) were lost; Coprosma lanceolaris—IBC was lost and in Atractocarpus stipularis—IBC was gained. 4. Discussion (a) Current patterns of isolation on Lord Howe Island Analysis of the genetic relatedness of individuals from 19 vas- (c) Population structure cular plant species on LHI confirms that environmental and Previous research has shown that no population structure has ecological gradients have roles in shaping the patterns of been observed within species in the SSG [26]. Here, we gene flow within this minute island. Environmental geography community altitude proximity to the coast 6 0.6 ************ * ** *** ** * ****** rstb.royalsocietypublishing.org 0.4

0.2

–0.2

–0.4 hl rn.R o.B Soc. R. Trans. Phil. –0.6

proximity to creeks soil water light soil pH 0.6 * * *** ** * ** ** * * * * * ** *** * 0.4 369

0.2 20130342 :

–0.2

–0.4

–0.6

slope topographic heterogeneity wind exposure 7 0.6 * * *** ** * * * * * * 6 0.4 5 0.2 4 0 3 –0.2 2 –0.4 1

–0.6 0 ACEG I KMOQS ACEG I KMOQS ACEG I KMOQS ACEG I KMOQS BDFHJ LNPR BDFH J LNPR BDFH J LNPR BDFH J LNPR Figure 2. Twelve matrix model averaging results of IBD, IBC and isolation by environmental variables (IBE). Each plot depicts effect sizes and 95% confidence intervals for each species grouped by environmental variable. See figure 1. A, Alyxia lindii; B, Alyxia ruscifolia; C, Atractocarpus stipularis; D, Coprosma prisca; E, Dracopyllum fitzgeraldii; F, Geniostoma petiolosum; G, Macropiper excelsum subsp. psittacorum; H, Macropiper hooglandii; I, Xylosma maidenii; J, Zygogynum howeanum; K, Coprosma huttoniana; L, Coprosma lanceolaris; M, Coprosma putida-N; N, Coprosma putida-S; O, Coprosma sp. nov.; P, Howea belmoreana; Q,Howea forsteriana; R, Metrosideros nervulosa; S, Metrosideros sclerocarpa. dissimilarity, as a single metric or included as specific environ- selection for genotypes with greater fitness in different habitats mental gradients, contributes to reductions in migration and/ [1,6,47–49] or habitat induced variation in assortative mating or reproduction in as many as 17 species, belonging to a range (e.g. through shifts in flowering time) are the most likely dri- of plant families and with varying degrees of phylogenetic vers of such patterns [50,51]. Whatever the mechanism, it is relatedness. A variety of environmental variables affect genetic clear that environmental variation plays a role in shaping the structure in the LHI species, with some species under the influ- genetic landscape within species even at fine scales. ence of single gradients and others subjected to many (figures 2 Despite the small size of the island, IBD was also a common and 3). The contributing factors to isolation within species are phenomenon, though less widespread than IBE in both assess- often different between sister species and congenerics (e.g. in ments. It is clear, therefore, that dispersal limitation can have Metrosideros, Macropiper and Coprosma). This points to the a significant impact on relatedness within plant taxa with a unpredictable and diverse ways in which plants can adapt mixture of dispersal abilities, even when the potential for geo- locally to their environment and is consistent with studies graphical isolation is severely limited. Although growth form that have shown adaptive responses of multiple plant species (i.e. plant habit) and dispersal mechanism have been impli- to environmental gradients [45,46]. The exact mechanisms cated in shaping patterns of spatial genetic structure in plants that lead to these relationships are less clear. However, natural [8,52,53], these factors have apparently little impact on the heterogeneous, these factors are not specific to LHI. Many 7 oceanic islands possess similar ranges of potential selection rstb.royalsocietypublishing.org pressures and anagenetic evolution of plant species on islands has been shown to decline with increasing heterogeneity geographycommunityaltitude coast creek soil waterlight soil pH slope heterogeneitywind * * * Al. lind. [27,56]. Such variability is equally common in continental set- * Al. rusc. tings, leading to observations of adaptation to environmental * * At. stip. gradients and IBE in a range of continental plant species * * * * * * Co. pris. [18,46]. It may also be possible that plants that have the ability Dr. fitz. to colonize isolated volcanic islands retain a high diversity * * * Ge. peti. of adaptive standing genetic variation [57]. However, there is * * Ma. exce. no direct evidence of this and island populations tend to

* * * B Soc. R. Trans. Phil. Ma. hoog. have lower genetic diversity than continental populations * * * * Xy. maid. [58]. It is more likely that the relaxed competition afforded by * Zy. howe. * * * * Co. hutt. newly emergent islands allows new niches to be exploited by * * * * * Co. lanc. colonizers, free from the crowded and highly competitive com- * * * Co. put-N. munities found in other locations. Through such ecological * Co. put-S. release, island species may occupy broader ecological niches * * * Co. sp. nv. [59,60]. Alleles that confer local adaptation in an island context 369

* * * * * * * Ho. belm. may not be sufficient to allow survival in similar, but more 20130342 : * * * * * * * Ho. fors. competitive, settings, giving rise to the prevalence of IBE on * * * Me. nerv. LHI. This is difficult to discount without examination of IBE * * ** * Me. scle. in species that are distributed both on islands and elsewhere. Figure 3. Heat map of IBD, IBC and isolation by environmental variables Investigation of IBE and IBD across an Antillean commu- (IBE). Negative effects in blue, positive effects in red. Asterisk denotes nity of lizards (Anolis) revealed similar variability in both significance (as in figure 1). IBD and IBE across different species to that observed in the LHI flora [17]. Converse to our findings, IBD was more common than IBE in Anolis species, suggesting that this may observed patterns within LHI species (table 1). These results not necessarily be a general feature of island taxa. However, imply that the small scales at which plants undergo reductions the comparison between LHI and the considerably larger in gene flow and subsequent speciation [54] are not only Antillean islands should be made with caution. When IBE is affected by dispersal limitation, but also reflect the selective tested across scales well beyond the dispersal distance of process that plant populations are subjected to in spatially the organism, gene flow can not only be directly affected structured environments. by the dissimilarity between the two sampled sites, but must Our assessment also included an estimate of isolation due to also depend on the intervening habitats. As a result, the indi- differences in the local plant community (IBC). These data were vidual-based, fine scale patterns observed here may not extracted from the literature and may be less accurate than direct translate to larger scale, population genetic isolation. Recent field observations. Despite this, the occurrence of IBC in several reviews that have examined the prevalence of IBE collated species suggests that this index reflects the importance of data in population-level studies that commonly used FST as a ecological relationships and acknowledges that competitive measure of gene flow, rather than the individual-based interactions between species can drive divergence [26,55]. approach exploited here [15,18]. Scale effects were not evident Importantly, in the 12 matrix analyses, IBC was only detected in these studies, but further examination of IBE and IBD at the in the SSG. Competitive interactions within these genera have individual level is required to establish whether these been implicated in speciation and the maintenance of species approaches produce consistent patterns. boundaries [26]. This result indicates that IBC may be the extra component of isolation necessary for local adaptation to progress to speciation in sympatry or parapatry. (c) Methodological considerations Intuitively, the simultaneous evaluation of the isolating effects of multiple variables is an improvement over other methods. (b) How common is isolation by environment? Amalgamation of the environmental variables can have an In general, the high frequency of IBE relationships observed effect in several ways, including the loss of the signal of iso- on LHI is consistent with that reported in the literature across lation for specific variables due to confounding effects. taxa [15–18]. Surprisingly, given the known adaptive ability Variables that are apparently responsible for the most isolation of many plants [45,46], plant studies generally demonstrate are not necessarily those that vary the most in the sample, and, lower IBE effect sizes [15], and the majority reject hypotheses as a result, the use of a principal component analysis to of IBE and detect IBD more frequently [18]. Why then is generate a dissimilarity matrix may mask the effect of there such a strong indication of IBE on LHI? The island may these variables. This is a particularly acute problem for those be unusual in that the highly variable structure of habitat species in which one strong association or a few weak associ- and environments can impose a great diversity of selection ations are present (e.g. Coprosma sp. nov., Coprosma putida-S, pressures. For LHI plants, the range of potential plant stressors, Z. howeanum). Similarly, combining variables that possess a such as restricted light in mountainous areas, salt exposure, mixture of isolating influences and counter-gradient patterns variable water availability, and temperature and humidity into a single measure has the effect of cancelling out any variability may induce strong selective environments that signal (e.g. C. huttoniana and C. prisca). As an extension of drive adaptation. Although the island environment is highly this, failing to estimate effects of individual variables through N Alyxia lindii S N Atractocarpus stipularis S 8 rstb.royalsocietypublishing.org

N Coprosma prisca S N Dracophyllum fitzgeraldii S hl rn.R o.B Soc. R. Trans. Phil.

N Geniostoma petiolosum S N Xylosma maidenii S 369 20130342 :

NSNMacropiper excelsum Macropiper hooglandii

Figure 4. Population genetic structure in the ASG species. Clusters within species were spatially intermixed, except in X. maidenii. Each vertical bar represents an individual and colours denote population membership. Individuals are ordered by latitude, N, north, S, south. Each plot depicts a STRUCTURE analysis with K set to 2.

the use of the single metric can lead to apparently spurious of individual loci with ecological variation (an indication of effects of IBD and IBC (D. fitzgeraldii, Coprosma sp. nov., local adaptation within species), (iv) ecological divergence Atractocarpus stipularis). On the other hand, when environ- of species in each genus and (v) competitive exclusion of con- mental effects may be individually weak or sources of generic species [8,26,28,30]. The implication from the current selection multivariate, combining the environmental variables study—that is, local adaptation within species on LHI is not into one measure of environmental dissimilarity may allow only possible, but the norm—further enhances the chances these patterns to be observed. This may explain the discre- that in some taxa this will lead to sufficiently strong repro- pancy between the two approaches used in this study, but ductive isolation to cause speciation. does not account for the differences in the frequency of IBE Distinct genetic population clusters were detected in one in plants found in this study and in the large number of studies species, and a further four species may also be dividing into gen- that have examined variables separately. Testing multiple vari- etic clusters, an indication that the flora harbours species at ables is much more likely to find correlation than testing only varying stages along the speciation-with-gene-flow continuum one. Comparison of our results with the assessments of iso- [3]. With the exception of D. fitzgeraldii,allofthesespecies lation by altitude and pH for the SSG of LHI plants (see [26]) demonstrated some evidence of both IBD and IBE. The partial demonstrates that restricting analyses to one or two variables spatial separation of populations of X. maidenii and presence will often miss the pattern that is evident when many are of IBD in this species do suggest geography has played a role taken into account. After correcting for geographical distance in the reduction of gene flow leading to population divergence. (partial Mantel test), the previous study found some evidence Nevertheless, the populations are divided between the wet of IBE in both Metrosideros species, both Howea species and south of the island and the dry north leading to strong patterns C. putida-N. The current analysis showed IBE in all of the of IBE, with hybrid individuals in both areas of the island. As a Metrosideros, Howea and Coprosma species. result, the relative influences of IBD and IBE on population divergence remain unclear. However, it is important to note that in all species except for X. maidenii the population clusters (d) New insights for speciation on Lord Howe Island are spatially intermixed, suggesting that the geographical The prevalence of environmental influences on relatedness component of isolation alone is not strong enough to cause within taxa supports previous research which concluded divergence. Again, this corroborates data suggesting that pre- that ecological speciation with gene flow may have occurred zygotic barriers (geographical isolation and flowering time multiple times on LHI in distantly related taxa (Howea, Metro- isolation) were not sufficiently strong to cause speciation in sideros and Coprosma). Evidence for ecological speciation in Metrosideros, Howea or the Coprosma radiation [26]. these three genera includes (i) divergence without significant IBD and IBE are clearly important phenomena for the geographical isolation, (ii) genetic signatures of divergent flora of LHI. Simultaneously estimating the isolating effects selection (detected using outlier analyses), (iii) associations of multiple environmental gradients provides a more detailed understanding of isolating barriers. Our analyses suggest that IBE in many taxa and locations suggests that ecologically 9 using this approach in other systems will reveal that IBE is driven isolation is, indeed, a major force in the accumulation rstb.royalsocietypublishing.org more pervasive than imagined. The apparently widespread of species diversity. occurrence of local adaptation supports a growing body of evidence for the potential for natural selection to overcome the homogenizing influence of gene flow, even at fine Acknowledgements. We thank Alan Walker, Jun Yim Ling, Rhian Smith, scales. Different plant taxa can respond to a variety of selec- Andrew Helmstetter, Hank and Sue Bower, Christo Haselden, Larry tion pressures and in some cases the strength of the ecological Wilson, the LHI board for their help with the research; the Royal Society, NERC and the ERC for funding; and the Editors, Daniel isolating barrier can lead to speciation in the same geographi- Crawford, Jonathan Wendel, Jeff Doyle, Pam and Doug Soltis, for cal area. Although it is possible that these patterns are unique giving us the opportunity to celebrate Gottlieb’s career and pioneer to the flora of LHI, the growing body of evidence supporting work on sympatric speciation (Am. J. Bot. 1973, 60, 545). hl rn.R o.B Soc. R. Trans. Phil.

References

1. Schluter D. 2000 The ecology of adaptive radiation. 15. Shafer ABA, Wolf JBW. 2013 Widespread evidence 26. Papadopulos AST, Price Z, Devaux C, Hipperson H,

Oxford, UK: Oxford University Press. for incipient ecological speciation: a meta-analysis Smadja CM, Hutton I, Baker WJ, Butlin RK, 369 2. Schluter D. 2009 Evidence for ecological speciation of isolation-by-ecology. Ecol. Lett. 16, 940–950. Savolainen V. 2013 A comparative analysis of the 20130342 : and its alternative. Science 323, 737–741. (doi:10. (doi:10.1111/ele.12120) mechanisms underlying speciation on Lord Howe 1126/science.1160006) 16. Wang IJ. 2013 Examining the full effects of Island. J. Evol. Biol. 26, 733–745. (doi: 10.1111/ 3. Feder JL, Egan SP, Nosil P. 2012 The genomics of landscape heterogeneity on spatial genetic jeb.12071) speciation-with-gene-flow. Trends Genet. 28, variation: a multiple matrix regression approach for 27. Papadopulos AST, Baker WJ, Savolainen V. 2013 342–350. (doi:10.1016/j.tig.2012.03.009) quantifying geographic and ecological isolation. Sympatric speciation of island plants: the natural 4. Nosil P. 2008 Speciation with gene flow could be Evolution 67, 3403–3411. (doi:10.1111/evo.12134) laboratory of Lord Howe island. In Speciation: common. Mol. Ecol. 17, 2103–2106. (doi:10.1111/j. 17. Wang IJ, Glor RE, Losos JB. 2013 Quantifying the natural processes, genetics and biodiversity 1365-294X.2008.03715.x) roles of ecology and geography in spatial genetic (ed. P Michalak). New York, NY: Nova Science. 5. Doebeli M, Dieckmann U, Metz JA, Tautz D. 2005 divergence. Ecol. Lett. 16, 175–182. (doi:10.1111/ 28. Savolainen V et al. 2006 Sympatric speciation in What we have also learned: adaptive speciation is ele.12025) palms on an oceanic island. Nature 441, 210–213. theoretically plausible. Evolution 59, 691–695. 18. Sexton JP, Hangartner SB, Hoffmann AA. 2014 (doi:10.1038/nature04566) 6. Rundle HD, Nosil P. 2005 Ecological speciation. Genetic isolation by environment or distance: which 29. Pickard J. 1983 Vegetation of Lord Howe Island. Ecol. Lett. 8, 336. (doi:10.1111/j.1461-0248.2004. pattern of gene flow is most common? Evolution Cunninghamia 1, 133–265. 00715.x) 68, 1–15. (doi:10.1111/evo.12258) 30. Babik W et al. 2009 How sympatric is speciation in 7. Funk DJ, Nosil P, Etges WJ. 2006 Ecological 19. Macarthur R, Levins R. 1967 The limiting similarity, the Howea palms of Lord Howe Island? Mol. Ecol. divergence exhibits consistently positive associations convergence, and divergence of coexisting species. 18, 3629–3638. (doi:10.1111/j.1365-294X.2009. with reproductive isolation across disparate taxa. Am. Nat. 101, 377–385. (doi:10.2307/2459090) 04306.x) Proc. Natl Acad. Sci. USA 103, 3209–3213. (doi:10. 20. Sork VL, Aitken SN, Dyer RJ, Eckert AJ, Legendre P, 31. Gavrilets S, Vose A. 2007 Case studies and 1073/pnas.0508653103) Neale DB. 2013 Putting the landscape into the mathematical models of ecological speciation. 2. 8. Papadopulos AST, Baker WJ, Crayn D, Butlin RK, genomics of trees: approaches for understanding Palms on an oceanic island. Mol. Ecol. 16, Kynast RG, Hutton I, Savolainen V. 2011 Speciation local adaptation and population responses to 2910–2921. (doi:10.1111/j.1365-294X.2007.03304.x) with gene flow on Lord Howe Island. Proc. Natl changing climate. Tree Genet. Genomes 9, 32. Vos P et al. 1995 AFLP: a new technique for DNA Acad. Sci. USA 108, 13 188–13 193. (doi:10.1073/ 901–911. (doi:10.1007/s11295-013-0596-x) fingerprinting. Nucleic Acids Res. 23, 4407–4414. pnas.1106085108) 21. Funk DJ, Egan SP, Nosil P. 2011 Isolation by (doi:10.1093/nar/23.21.4407) 9. Butlin RK, Galindo J, Grahame JW. 2008 Sympatric, adaptation in Neochlamisus leaf beetles: host- 33. Green PS. 1994 Flora of Australia volume 49 oceanic parapatric or allopatric: the most important way to related selection promotes neutral genomic islands 1. Canberra, Australia: Australian classify speciation?. Phil. Trans. R. Soc. B 363, divergence. Mol. Ecol. 20, 4671–4682. (doi:10. Government Publishing Service. 2997–3007. (doi:10.1098/rstb.2008.0076) 1111/j.1365-294X.2011.05311.x) 34. Sappington JM, Longshore KM, Thompson DB. 2007 10. Coyne JA. 2011 Speciation in a small space. Proc. 22. Nosil P, Egan SP, Funk DJ. 2008 Heterogeneous Quantifying landscape ruggedness for animal Natl Acad. Sci. USA 108, 12 975–12 976. (doi:10. genomic differentiation between walking-stick habitat analysis: a case study using bighorn sheep 1073/pnas.1110061108) ecotypes: ‘isolation-by-adaptation’ and multiple in the Mojave Desert. J. Wildl. Manage. 71, 11. Wright S. 1943 Isolation by distance. Genetics roles for divergent selection. Evolution 62, 1419–1426. (doi:10.2193/2005-723) 28, 114. 316–336. (doi:10.1111/j.1558-5646.2007.00299.x) 35. Doyle JJ, Doyle JL. 1987 A rapid DNA isolation 12. Jenkins DG et al. 2010 A meta-analysis of isolation 23. Manel S, Holderegger R. 2013 Ten years of procedure for small amounts of fresh leaf tissue. by distance: relic or reference standard for landscape genetics. Trends Ecol. Evol. 28, 614–621. Phytochem. Bull. 19, 11–15. landscape genetics? Ecography 33, 315–320. (doi:10.1016/j.tree.2013.05.012) 36. Csiba L, Powell MP. 2006 DNA extraction protocols. (doi:10.1111/j.1600-0587.2010.06285.x) 24. Burnham KP, Anderson DR. 2002 Model selection In DNA and tissue banking for biodiversity 13. Nosil P. 2012 Ecological speciation. Oxford, UK: and multimodel inference: a practical information- and conservation: theory, practice and uses Oxford University Press. theoretic approach. New York, NY: Springer. (eds V Savolainen, MP Powell, K Davis, G Reeves, 14. Crispo E, Bentzen P, Reznick DN, Kinnison MT, 25. Grueber CE, Nakagawa S, Laws RJ, Jamieson IG. A Corthals), pp. 48–51. Richmond, Surrey, UK: Hendry AP. 2006 The relative influence of natural 2011 Multimodel inference in ecology and Royal Botanic Gardens, Kew. selection and geography on gene flow in guppies. evolution: challenges and solutions. J. Evol. Biol. 37. R Development Core Team. 2009 R: a language and Mol. Ecol. 15, 49–62. (doi:10.1111/j.1365-294X. 24, 699–711. (doi:10.1111/j.1420-9101.2010. environment for statistical computing. Vienna, 2005.02764.x) 02210.x) Austria: R Foundation for Statistical Computing. 38. Strimmer K. 2008 fdrtool: a versatile R package for environmental gradient. Oecologia 170, 89–99. populations. Mol. Ecol. 13, 921–935. (doi:10.1046/ 10 estimating local and tail area-based false discovery (doi:10.1007/s00442-012-2291-2) j.1365-294X.2004.02076.x) rstb.royalsocietypublishing.org rates. Bioinformatics 24, 1461–1462. (doi:10.1093/ 46. Clausen J, Keck DD, Hiesey WM. 1940 Experimental 53. Loveless MD, Hamrick JL. 1984 Ecological bioinformatics/btn209) studies on the nature of species. I. Effect of varied determinants of genetic structure in plant 39. Hardy OJ, Vekemans X. 2002 SPAGeDi: a versatile environments on western North American plants. populations. Annu. Rev. Ecol. Syst. 15, 65–95. computer program to analyse spatial genetic structure Washington, DC: Carnegie Institution of (doi:10.2307/2096943) at the individual or population levels. Mol. Ecol. Notes Washington. 54. Kisel Y, Barraclough TG. 2010 Speciation has a 2, 618. (doi:10.1046/j.1471-8286.2002.00305.x) 47. Rundle HD, Whitlock MC. 2001 A genetic spatial scale that depends on levels of gene flow. 40. Ehrich D. 2006 aflpdat: a collection of R functions interpretation of ecologically dependent isolation. Am. Nat. 175, 316–334. (doi:10.1086/650369) for convenient handling of AFLP data. Mol. Ecol. Evolution 55, 198–201. (doi:10.2307/2640701) 55. Schluter D. 1994 Experimental evidence that Notes 6, 603–604. (doi:10.1111/j.1471-8286.2006. 48. Rice WR. 1987 Speciation via habitat specialization: competition promotes divergence in adaptive

01380.x) the evolution of reproductive isolation as a radiation. Science 266, 798–801. (doi:10.1126/ B Soc. R. Trans. Phil. 41. Fox J, Weisberg S. 2011 An R companion to applied correlated character. Evol. Ecol. 1, 301–314. (doi:10. science.266.5186.798) regression, 2nd edn. Thousand Oaks, CA: Sage. 1007/bf02071555) 56. Stuessy TF, Jakubowsky G, Go´mez RS, Pfosser M, 42. Pritchard JK, Stephens M, Donnelly P. 2000 49. Via S, Bouck AC, Skillman S. 2000 Reproductive isolation Schlu¨ter PM, Fer T, Sun BY, Kato H. 2006 Anagenetic Inference of population structure using multilocus between divergent races of pea aphids on two hosts. II. evolution in island plants. J. Biogeogr. 33, 1259– genotype data. Genetics 155, 945–959. selection against migrants and hybrids in the parental 1265. (doi:10.1111/j.1365-2699.2006.01504.x)

43. Falush D, Stephens M, Pritchard JK. 2007 Inference environments. Evolution 54, 1626–1637. (doi:10.1111/ 57. Barrett RDH, Schluter D. 2008 Adaptation from 369

of population structure using multilocus genotype j.0014-3820.2000.tb00707.x) standing genetic variation. Trends Ecol. Evol. 23, 20130342 : data: dominant markers and null alleles. Mol. Ecol. 50. Devaux C, Lande R. 2008 Incipient allochronic 38–44. (doi:10.1016/j.tree.2007.09.008) Notes 7, 574–578. (doi:10.1111/j.1471-8286.2007. speciation due to non-selective assortative mating 58. Frankham R. 1997 Do island populations have less 01758.x) by flowering time, mutation and genetic drift. genetic variation than mainland populations? 44. Evanno G, Regnaut S, Goudet J. 2005 Detecting the Proc. R. Soc. B 275, 2723–2732. (doi:10.1098/rspb. Heredity 78, 311–327. (doi:10.1038/hdy.1997.46) number of clusters of individuals using the software 2008.0882) 59. Wilson EO. 1961 The nature of the taxon cycle in STRUCTURE: a simulation study. Mol. Ecol. 14, 51. Weissing F, Edelaar P, Doorn GS. 2011 Adaptive the Melanesian ant fauna. Am. Nat. 95, 169–193. 2611–2620. (doi:10.1111/j.1365-294X.2005. speciation theory: a conceptual review. Behav. Ecol. (doi:10.2307/2458389) 02553.x) Sociobiol. 65, 461–480. (doi:10.1007/s00265-010- 60. Lister BC. 1976 The nature of niche expansion in 45. Haider S, Kueffer C, Edwards P, Alexander J. 2012 1125-7) West Indian Anolis lizards I: ecological consequences Genetically based differentiation in growth of 52. Vekemans X, Hardy OJ. 2004 New insights from of reduced competition. Evolution 30, 659–676. multiple non-native plant species along a steep fine-scale spatial genetic structure analyses in plant (doi:10.2307/2407808)